Self-Supervised Learning refers to supervised machine learning techniques where labels (which are required for supervised learning) don't need to be created by humans. That is the case, for example, in the task of predicting the next word in a sentence, or a missing patch of an image. Often times these self-supervised techniques are used to pre-train models which can be later fine-tuned on similar tasks -- e.g. a model which predicts the next word in a sentence can be fine-tuned (on a dataset of human-provided labels) into a question-answering model.